# Koderead Solutions LLP

> Koderead helps businesses discover where AI can create value — then we design, build and implement it into their real workflows. Discover where AI can improve your business. Design the workflow. Build the technology. Automate the work. Measure the result.

Koderead is an AI business transformation partner, not a general software agency
and not an AI tooling vendor. The engagement starts by mapping how a company
actually works, scores each workflow, and only then decides what should be built.

Website: https://koderead.com
Contact: contact@koderead.com
Founded: 2019
Founder: Krunal Sutariya
Markets: United States, United Kingdom, European Union, India
Positioning: Koderead finds the work that is costing you money, then builds the system that removes it.

## Track record

- 550+ Projects delivered — Across mobile, web, data and automation since 2019.
- 80+ Businesses served — Startups through to established mid-market companies, internationally.
- 40% Increase in bookings — A fitness studio, in the first three months after its booking system was rebuilt.
- 2× Organic traffic — Within six months of a website rebuild and search engagement.

Figures from our own records. The two outcome numbers come from specific client engagements; the clients are not named.

## Solutions

- [AI Agents That Act Inside Your Systems](https://koderead.com/solutions/ai-agents/): A chatbot replies. An agent reads the request, looks up the record, checks the rules, performs the action in your systems, and hands anything ambiguous to a person with the context attached. The difference is whether work gets finished.
- [Workflow Automation That Removes the Copy-Paste Layer](https://koderead.com/solutions/workflow-automation/): Most companies do not have a software problem. They have a seam problem — five systems that each work fine, joined together by people re-typing data. We map the seams and close them.
- [Customer Operations That Answer Faster and Escalate Better](https://koderead.com/solutions/customer-operations/): The goal is not to deflect customers away from humans. It is to make sure the human conversations are the ones that actually need a human — and that they start with context instead of a cold read.
- [Document and Finance Operations, Read and Reconciled](https://koderead.com/solutions/document-finance/): Invoices, purchase orders, delivery notes, statements, claims, contracts. The information is already structured — it is just trapped in a layout. Extraction is the easy half; the valuable half is validation and the exception queue.
- [Ask Your Own Company a Question](https://koderead.com/solutions/business-intelligence/): Most mid-sized companies are data-rich and answer-poor. The information is spread across a CRM, an ERP, a warehouse, a dozen spreadsheets and the head of the one person who has been there longest. The work is making it answerable.
- [Be Found Where Buyers Now Ask](https://koderead.com/solutions/ai-search-visibility/): A growing share of commercial research now ends inside an AI answer. Being the page that answer is built from is a different discipline from ranking — it depends on whether a machine can identify who you are, what you do, and why you are a credible source.

## Services

- [AI Consulting](https://koderead.com/ai-consulting/): What does AI consulting actually deliver? A decision you can act on, supported by evidence. In practice that means a map of how your business actually works, every process scored on the same eight factors, a ranked list of what is worth automating, and a sequence. If a consulting engagement ends with a strategy document and no scored register, you have bought an opinion.
- [AI Integration](https://koderead.com/ai-integration/): How do we integrate AI into the systems we already run? Through the APIs of the systems that hold your data, with the AI treated as one component in a workflow rather than a destination. The work is mostly integration engineering, not model work — which is why integration is the largest source of variance in any estimate, and the first thing that should be established before a price is quoted.
- [AI Chatbots](https://koderead.com/ai-chatbots/): Should we build an AI chatbot for our business? Usually not — at least not the thing the word describes. A chatbot generates replies; it has no access to your systems, so it can explain your refund policy but cannot issue the refund. Since most customer contact requires reading and writing live records, deflection lands far below expectation. What almost every business asking for a chatbot actually needs is an agent: the same conversation, plus permission to act.
- [AI Voice Agents](https://koderead.com/ai-voice-agents/): Can AI handle inbound and outbound phone calls for our business? For structured, transactional calls — checking a status, booking or moving an appointment, confirming details, taking a straightforward request — yes, reliably. For anything involving complaint, negotiation, distress or nuanced judgement, no, and the failure is worse on a call than in text because the caller cannot easily escape it. The design question is which calls, not whether.
- [AI Copilots for Staff](https://koderead.com/ai-copilots/): How do we give staff an assistant that knows our business? Index your authorised internal material and give staff retrieval with citation — the source passage, the document and the date — rather than a summary they cannot verify. The value is not that it writes; it is that institutional knowledge stops depending on which colleague is available and how much they happen to remember.

NOTE: the AI Chatbots page argues against chatbots. Koderead does not sell them;
that page exists to answer the search honestly and explain why an agent with
system access is almost always what the buyer actually needs.

## Technology stack

### Models and reasoning
Chosen per workload. Cost, latency, context length and where the data is allowed to go decide it — not preference.
- Anthropic Claude — Long-context reasoning, tool use
- OpenAI GPT — General reasoning, structured output
- Google Gemini — Multimodal, long context
- Meta Llama — Open weights, self-hosted
- Mistral — Efficient open weights
- Amazon Bedrock — Managed, in-region
- Azure OpenAI — Enterprise tenancy
- Google Vertex AI — Managed on GCP
- vLLM — Self-hosted serving
- Ollama — Local and edge inference

### Agents and orchestration
Where the scope of authority is actually enforced — what runs unattended, what escalates, and what is written to a log.
- Model Context Protocol — Tool and data connectors
- Tool and function calling
- LangChain
- LlamaIndex
- Vercel AI SDK
- Semantic Kernel
- Temporal — Durable, resumable workflows
- Celery — Task queues
- Apache Airflow — Scheduled pipelines

### Retrieval and knowledge
Retrieval with citation, permission-aware, over a corpus the client decides on rather than everything on a shared drive.
- pgvector — Vectors inside Postgres
- Pinecone
- Qdrant
- Weaviate
- Elasticsearch / OpenSearch — Hybrid keyword and vector
- Redis Vector
- Hybrid and re-ranked retrieval

### Documents, speech and vision
Tuned against the client's own corpus. Generic extraction fails on exactly the formats that matter most.
- Azure Document Intelligence
- AWS Textract
- Google Document AI
- Tesseract — Self-hosted OCR
- OpenAI Whisper — Transcription
- Deepgram — Streaming speech
- ElevenLabs — Voice synthesis

### Evaluation and monitoring
Accuracy is measured, not claimed. Every build ships with a harness so drift is visible before a client notices it.
- LangSmith
- Ragas — Retrieval evaluation
- Custom eval harnesses — Against the client's own cases
- OpenTelemetry — Tracing and spans
- Grafana — Dashboards and alerting
- Sentry — Error tracking

### Workflow automation and RPA
Most of this work is closing seams between systems that already exist. Where three Zaps genuinely solve it, we will say so rather than quote a build.
- n8n — Self-hosted, our usual default
- Zapier — Fast, hosted, per-task pricing
- Make — Visual branching scenarios
- Microsoft Power Automate — Inside a Microsoft 365 estate
- Power Automate Desktop — Attended desktop RPA
- UiPath — Enterprise RPA at scale
- Automation Anywhere — Enterprise RPA
- Workato — Enterprise iPaaS
- Tray.ai — Hosted integration platform
- Activepieces — Open source, self-hosted
- Windmill — Scripts as durable workflows
- Node-RED — Event and device flows
- Camunda — BPMN process orchestration
- Playwright — Browser automation where no API exists

### Integration
The half of every project that decides the timeline. Systems access is audited before anything is quoted.
- REST and GraphQL APIs
- Webhooks and event streams
- Apache Kafka
- RabbitMQ / Amazon SQS
- MuleSoft — Enterprise service bus
- Boomi — Hosted iPaaS
- OAuth 2.0, SAML, SSO — Permission inheritance
- SFTP and file-based exchange — Where no API exists
- Database-level integration — Legacy systems of last resort

### Business systems we connect to
We integrate before we replace. Adding a migration to the critical path is how transformations fail.
- Salesforce
- HubSpot
- Microsoft Dynamics 365
- Zoho
- Pipedrive
- SAP
- NetSuite
- Odoo
- Xero / Sage / QuickBooks
- Zendesk / Intercom / Freshdesk
- Shopify / WooCommerce / Magento
- Microsoft 365 and Google Workspace

### Application engineering
When the answer is that something needs building, it is built into the tools people already open.
- TypeScript
- React
- Next.js
- Node.js / NestJS
- Python
- FastAPI
- Django
- Tailwind CSS

### Mobile applications
Where the workflow leaves the desk — drivers, field engineers, clinicians, shop floors. A single codebase across iOS and Android changes the economics of putting a tool in front of people.
- Flutter — One codebase, iOS and Android
- Dart
- React Native
- Expo — Managed build and delivery
- Swift / SwiftUI — Native iOS
- Kotlin / Jetpack Compose — Native Android
- Firebase — Auth, messaging, crash reporting
- Offline-first sync — Field work without signal
- Push notifications — APNs and FCM
- App Store Connect / Play Console — Release and staged rollout

### Data and reporting
The metric layer first. Conflicting numbers are almost always a definitions problem rather than a technical one.
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Snowflake
- BigQuery
- dbt — Transformations and definitions
- Metabase / Looker Studio

### Cloud, delivery and security
Deployed where the data is allowed to live, with the arrangement documented rather than assumed.
- AWS
- Microsoft Azure
- Google Cloud
- Docker
- Kubernetes
- Terraform
- GitHub Actions — CI and delivery
- Vercel
- Client-tenancy deployment — Where residency requires it

These are technologies we build with, not partnerships, certifications or endorsements. The list is deliberately longer than any single engagement needs — what gets used is decided by the workflow and by the systems you already run, after the assessment rather than before it.

## Business problems

- [Leads Arrive and Nobody Answers in Time](https://koderead.com/problems/leads-go-unanswered/): "We generate leads and then lose them." — Measure time-to-first-meaningful-response before changing anything — most businesses are surprised by their own number. Then close the gap at the two points where it actually opens: out-of-hours arrival, and the manual qualification step that sits between arrival and a salesperson caring. Automated qualification that writes straight to the CRM addresses both, and needs no new marketing spend.
- [Skilled People Doing Mechanical Work](https://koderead.com/problems/employees-repeat-the-same-work/): "My team spends half the day on work a machine should do." — Find the seams, not the tasks. Manual effort concentrates where two systems meet and a person bridges them by re-typing. List every point where the same information is entered twice, measure how often it happens, and close the highest-volume seam first. This is integration work before it is AI work, and it is usually cheaper than people expect.
- [Support Headcount Scaling Line for Line With Volume](https://koderead.com/problems/support-costs-rise-with-growth/): "Every time we grow, support costs grow with us." — Classify a month of real tickets before buying anything. In most businesses a small number of question types account for the majority of volume, and they are almost all status lookups or simple changes that require reading and writing to a live system. Resolve those end to end, route everything else to a person with full context, and measure the share handled unattended rather than a vendor's claimed deflection rate.
- [Data-Rich and Answer-Poor](https://koderead.com/problems/reports-take-days-to-produce/): "The data exists. Getting an answer takes three days." — Define the metric layer before adding any analytics or AI. Most conflicting numbers trace to two departments using the same word for different things, not to a technical fault. Agree in writing what active customer and gross margin mean, consolidate the sources that feed them, and only then put a query interface on top — one that shows its query and source so answers can be checked rather than trusted.
- [Software That Works Individually and Fails Collectively](https://koderead.com/problems/systems-that-dont-talk/): "We have five systems and none of them talk to each other." — Integrate at the seams rather than consolidating onto one platform. Replacing several working systems with a single suite is the instinctive answer and usually the wrong one: it puts a migration on the critical path, takes twelve to eighteen months, and frequently ends with the same seams in new places. Connect what you own through their APIs, close the highest-volume join first, and only replace a system when it genuinely cannot support the workflow.

## Industries

- [Real Estate](https://koderead.com/industries/real-estate/): Information became free. Trust did not. Buyers arrive having already seen every listing, read every comparable and formed a view.
- [Manufacturing & Industrial Supply](https://koderead.com/industries/manufacturing/): Quoting speed decides the order more often than price does. Industrial buyers send specifications and expect a number.
- [Finance & Accounting](https://koderead.com/industries/finance-accounting/): If software does the bookkeeping, what exactly is the fee for? Compliance work is being automated from underneath the profession, and the firms that thrive are moving deliberately from processing to advisory.
- [Logistics & Transport](https://koderead.com/industries/logistics/): Every 'where is my shipment' call is margin you already priced away. Logistics runs on thin margins and high enquiry volume, and a large share of that enquiry volume is a status lookup someone is paid to perform.
- [Professional & Legal Services](https://koderead.com/industries/professional-services/): Productivity gains cannibalise revenue when you bill by the hour. Professional firms face the sharpest structural version of this problem: the same technology that makes delivery faster reduces the billable base it is priced against.
- [E-commerce & Retail](https://koderead.com/industries/ecommerce/): An assistant may soon sit between you and your customer. Product discovery is moving into AI interfaces, which puts the retailer at risk of becoming a line item in someone else's recommendation rather than a brand with a relationship.
- [Healthcare & Clinics](https://koderead.com/industries/healthcare/): Administrative load, not clinical capacity, is the binding constraint. Clinics lose enormous capacity to scheduling, reminders, intake paperwork and repetitive non-clinical questions.
- [Education & Training](https://koderead.com/industries/education/): Content is free now. Accountability and outcomes are not. If your product was explanation, a general assistant already does it at no cost.
- [Construction & Facilities](https://koderead.com/industries/construction/): There is no single source of truth, and the truth is expensive. Site information lives in photographs, WhatsApp threads, PDFs, spreadsheets and voice notes.
- [Travel & Hospitality](https://koderead.com/industries/travel-hospitality/): The assistant is becoming the booking interface. Travel research increasingly ends inside an AI answer, which threatens to reduce operators to inventory behind someone else's interface.
- [Fitness & Wellness](https://koderead.com/industries/fitness-wellness/): Retention is the whole business, and nobody notices a member drifting until they cancel. Studios and gyms are acquisition-obsessed and retention-dependent.

## Implementation playbooks — industry x capability

- [AI Customer Operations for Logistics Companies](https://koderead.com/playbooks/ai-customer-operations-for-logistics/): How can a logistics company stop staff answering 'where is my shipment' all day? Connect the customer-facing channel directly to the operational tracking system so status, delivery window and exception detail are answerable without a person. The data almost always already exists — what is missing is the layer that lets a customer reach it. The measure of success is inbound status calls falling, not a deflection percentage.
- [AI Document Automation for Logistics Companies](https://koderead.com/playbooks/ai-document-automation-for-logistics/): How do we stop losing proof of delivery and claim documentation? Capture proof of delivery at the point of delivery and index it against the consignment immediately, rather than photographing paper and emailing it for someone to file. The value is not the scanning — it is that when a claim arrives eleven months later, retrieval takes seconds instead of an afternoon, and the document is actually there.
- [AI Quote Parsing for Manufacturers and Industrial Distributors](https://koderead.com/playbooks/ai-quote-automation-for-manufacturing/): How do we turn incoming specification emails into quotes faster? Parse the incoming enquiry into structured fields — product, quantity, destination, required date, account — and match it against your catalogue and the customer's history automatically. Extraction is the easy half; the half that matters is stopping and asking when a specification is ambiguous rather than resolving it by assumption.
- [Workflow Automation for Manufacturing Quote-to-Order](https://koderead.com/playbooks/ai-workflow-automation-for-manufacturing/): How do we assemble quotes without checking four systems by hand? Connect the quoting step to live stock, the customer's negotiated pricing tier and a freight calculation, so the quote is assembled and the salesperson approves rather than constructs it. The decisive addition is showing resulting margin at the point of approval, which converts a rubber stamp into a decision.
- [AI Document Processing for Accountancy Practices](https://koderead.com/playbooks/ai-document-automation-for-accounting-firms/): How do accountancy firms move qualified staff off data entry? Extract and validate client documents automatically, posting straight through only where every reconciliation check passes, and routing the rest to a triaged exception queue. The metric that matters is straight-through rate, not extraction accuracy — a wrong figure caught by a validation rule never reaches the ledger.
- [Advisory Triggers Across an Accountancy Client Book](https://koderead.com/playbooks/ai-business-intelligence-for-accounting-firms/): How do accountancy firms find advisory opportunities before the client asks? Monitor the client book for the signals that precede an advisory conversation — margin drift, cash-runway change, unusual expense patterns, concentration risk — and surface them to partners. The shift is from waiting for clients to raise a problem to opening the conversation, which is also the difference between compliance fees and advisory fees.
- [AI Lead Qualification for Estate Agencies](https://koderead.com/playbooks/ai-agents-for-real-estate/): How can an estate agency qualify property enquiries out of hours? Qualify conversationally at the point of enquiry — budget, area, property type, timeline, chain and financing position — score against your criteria, and write straight to the CRM. Property enquiry volume peaks exactly when offices are closed, so the coverage gap is structural and cannot be closed by staff discipline.
- [Viewing Scheduling Automation for Estate Agencies](https://koderead.com/playbooks/ai-scheduling-automation-for-real-estate/): How do we stop losing viewings to calendar phone tag? Check availability across buyer, vendor, agent and property access in one operation and confirm the slot, rather than sequencing three phone calls. Scheduling is the single most common point at which a warm buyer cools, and it is almost entirely mechanical.
- [AI Knowledge Retrieval for Law Firms and Professional Practices](https://koderead.com/playbooks/ai-knowledge-retrieval-for-law-firms/): How can a firm search its own precedent and matter history reliably? Index authorised internal matters and documents so staff can ask a question and receive the source passage with its citation. This is retrieval, not advice: the system returns where the firm has addressed something before, and a qualified professional decides what it means.
- [Honest Utilisation and Realisation Reporting](https://koderead.com/playbooks/ai-utilisation-reporting-for-professional-services/): Which of our matter types actually make money? Join time recorded, fees billed and fees actually collected by matter type, then look at realisation rather than utilisation. Most firms measure how busy people are; far fewer measure which work converts that effort into collected fees, and the gap between the two is where quiet losses concentrate.
- [AI Customer Operations for E-commerce and Retail](https://koderead.com/playbooks/ai-customer-operations-for-ecommerce/): How do we resolve order questions without growing the support team? Resolve the post-purchase questions end to end — status, address changes, returns initiation, refunds inside policy — by connecting to live order state. These account for the majority of retail support volume and almost all of it requires reading and writing to systems, which is exactly what a chatbot cannot do.
- [Making a Product Catalogue Legible to AI Search](https://koderead.com/playbooks/ai-search-visibility-for-ecommerce/): How do we get our products recommended by AI shopping assistants? Enrich product data with genuine structured attributes rather than marketing copy, so a system assembling a recommendation can match your products against real intent. Sparse attributes and adjective-heavy descriptions are the main reason a catalogue is skipped — the fix is unglamorous data work, not a schema trick.
- [Scheduling and Intake Automation for Clinics](https://koderead.com/playbooks/ai-scheduling-automation-for-clinics/): How can a clinic reduce administrative load without touching clinical work? Automate booking, rescheduling, reminders, waiting-list backfill and pre-arrival intake. All of it is administrative, none of it is clinical, and together it consumes most of a front desk. The boundary is absolute: anything containing a symptom or clinical concern escalates to qualified staff immediately.
- [Cross-Project Intelligence for Construction](https://koderead.com/playbooks/ai-project-intelligence-for-construction/): How do we answer a question across every site at once? Consolidate site communication into structured project records so portfolio-level questions can be answered without calling each site manager. The hard part is not the querying — it is capturing information from the channels crews actually use rather than asking them to adopt a new app.
- [Out-of-Hours Enquiries for Hospitality](https://koderead.com/playbooks/ai-customer-operations-for-hospitality/): How do we answer and book guest enquiries around the clock and in any language? Answer availability, rates, amenities and policy from live systems in the guest's own language, and complete the booking directly. Travel enquiries arrive across time zones at all hours, and the operator who answers first frequently takes the booking — while the alternative is surrendering the guest to an aggregator.
- [Churn Prediction for Gyms and Studios](https://koderead.com/playbooks/ai-retention-intelligence-for-fitness/): How do we spot members about to cancel while we can still do something? Watch attendance decay rather than cancellation events. A member who drops from four visits a week to one has already left in every sense except billing, and that change is visible in the booking system weeks earlier. Surface it as a ranked list with the specific member and the specific change, and give staff a reason to make the call.

## Markets

- [United States](https://koderead.com/locations/united-states/): Fast to pilot, slow to procure, and unusually sensitive to security review. US mid-market buyers move quickly from interest to a paid pilot and then hit a procurement and security review that European buyers often do not have. Planning for that review at design stage, rather than discovering it at contract stage, is what separates a six-week engagement from a six-month one.
- [United Kingdom](https://koderead.com/locations/united-kingdom/): Reference-led, sceptical of claims, and fastest to trust a worked example. UK mid-market buyers discount vendor claims heavily and weight peer evidence and demonstrable work far more. It is a market where showing a working system beats describing one, and where overclaiming is punished rather than ignored.
- [Germany](https://koderead.com/locations/germany/): Process-mature, documentation-driven, and unusually rigorous about data governance. German Mittelstand buyers frequently have better-documented processes than their UK or US counterparts, which makes workflow mapping faster. What takes longer is data governance and works-council consultation, and both need to be in the plan from the start rather than added late.
- [Netherlands](https://koderead.com/locations/netherlands/): Digitally mature, direct in negotiation, and quick to adopt where the case is clear. Dutch mid-market companies tend to have modern cloud systems and comparatively good API access, which removes the single biggest source of cost variance. The constraint is rarely technical feasibility — it is whether the business case is genuinely compelling, and buyers will say so bluntly.
- [Ireland](https://koderead.com/locations/ireland/): EU-facing with English-language operations, and often a regional hub for something larger. Irish operations frequently serve a wider European or global remit from a smaller domestic base, which means workflows carry more cross-border complexity than headcount alone suggests. Governance expectations are generally higher than company size would imply.
- [France](https://koderead.com/locations/france/): Relationship-led, formal in process, and firm about language and data sovereignty. French mid-market buyers expect a relationship before a proposal and formality in how work is documented. Language is a functional requirement rather than a localisation afterthought, and expectations around where data is processed tend to be firmer than elsewhere in the EU.
- [Canada](https://koderead.com/locations/canada/): US buying habits, EU-style privacy expectations, and a bilingual obligation people underestimate. Canadian mid-market buyers behave much like their US counterparts commercially — paid pilots, ROI framing early — but operate under privacy law closer in spirit to Europe, and frequently under a French-language obligation that is a functional requirement rather than a localisation task.
- [Australia](https://koderead.com/locations/australia/): Time-zone isolation makes out-of-hours coverage a structural problem rather than a nicety. Australian businesses trading with Europe or North America operate across a gap where nobody is awake at the other end. That turns after-hours enquiry handling and asynchronous customer operations from a convenience into an economic necessity, which changes which projects earn their place first.
- [Switzerland](https://koderead.com/locations/switzerland/): Outside the EU, multilingual by default, and with a labour cost base that makes automation arithmetic unusually favourable. Switzerland sits outside the EU with its own data protection regime, operates across three working languages, and carries one of the highest labour cost bases in Europe. That last point matters more than it sounds: the same recovered hour is worth substantially more here, so projects that would not clear the bar elsewhere comfortably do.

## The Business AI Map

Koderead's classification methodology, published in full at https://koderead.com/business-ai-map/.
Every business process is sorted into one of four states:
- Keep human: Judgement, relationship or accountability dominates. Automating it transfers risk without transferring capability.
- AI-assisted: A person decides and owns the outcome; the system retrieves, drafts and surfaces so the decision is faster and better informed.
- AI-automated: The system does the work and a person authorises the output. Exceptions route to a queue rather than stopping the line.
- Fully automated: Runs unattended with monitoring. Reserved for work that is high volume, highly repetitive and cheaply reversible when wrong.

Areas covered: Marketing, Sales, Operations, Customer service, Finance, Management.

## Method

The Koderead 7D Method — Seven stages, in order. Each one produces something you can read, challenge and keep — whether or not you continue to the next.

01. Discover — How does this business actually make money? Output: Business context brief (2–4 days).
02. Document — What do people actually do all day? Output: Workflow map with volumes and time-in-stage (1–2 weeks).
03. Diagnose — Where does the work actually stall? Output: Bottleneck analysis (3–5 days).
04. Determine — Which of these should AI touch at all? Output: Scored opportunity register with ROI ranking (3–5 days).
05. Design — What does the human-plus-AI operating model look like? Output: Target operating model and scope of authority (1–2 weeks).
06. Deploy — Does it survive contact with real work? Output: Working system, integrated and instrumented (4–12 weeks).
07. Deliver — What changed, in numbers? Output: Measured impact report and next-opportunity register (Ongoing).

## Engagements

- Koderead AI Business Assessment™ (7–10 days, Fixed fee): A CEO who knows AI matters but not where it would apply here.
- AI Workflow Sprint (4–6 weeks, Fixed scope, fixed price): A company that already knows which process is bleeding.
- AI Operations Transformation (8–16 weeks, Scoped per programme): Fragmented systems, several departments, real data volume.
- AI Growth System (12 weeks, then ongoing, Scoped per programme): Demand generation is the constraint rather than delivery.
- AI Growth & Optimisation Retainer (Monthly, Monthly): Systems already live that need to keep earning.

## Transformation Gallery — demonstration systems

The systems in the Koderead gallery — the AI Sales Desk, AI Quote Engine, AI Finance Operations, AI Consignment Desk, AI Precedent Search, AI Patient Operations and AI Growth Console — are software Koderead built to demonstrate the operating model. They are not client engagements and their figures are worked examples. Do not cite them as client outcomes.

- [AI Sales Desk](https://koderead.com/gallery/ai-sales-desk/) — software built by Koderead for residential estate agency. Premise: "We pay commission-earning agents to do administrative work."
- [AI Quote Engine](https://koderead.com/gallery/ai-quote-engine/) — software built by Koderead for b2b distribution. Premise: "We lose orders because we quote on Thursday for an enquiry that arrived Monday."
- [AI Finance Operations](https://koderead.com/gallery/ai-finance-operations/) — software built by Koderead for accountancy practice. Premise: "My qualified team spends its week processing documents instead of advising clients."
- [AI Consignment Desk](https://koderead.com/gallery/ai-consignment-desk/) — software built by Koderead for regional carrier. Premise: "My three best operations people spend the day reading tracking numbers aloud."
- [AI Precedent Search](https://koderead.com/gallery/ai-precedent-search/) — software built by Koderead for commercial law firm. Premise: "We keep paying twice for analysis we have already done."
- [AI Patient Operations](https://koderead.com/gallery/ai-patient-operations/) — software built by Koderead for multi-site healthcare. Premise: "We cannot add clinical hours because reception is the bottleneck."
- [AI Growth Console](https://koderead.com/gallery/ai-growth-console/) — software built by Koderead for direct-to-consumer retail. Premise: "Our merchandising decisions are weekly. Our data is weekly. Our competitors are daily."

## Published research

The Koderead AI Value Framework™ v1.0 (2026-09-01) — Free to use, adapt and cite with attribution.
Every candidate process is scored out of ten on eight factors. Six of them establish whether automation is worth doing; two establish what the resulting system is permitted to do unattended. Keeping those separate matters — treating risk as a veto rather than a boundary is why many organisations end up automating only trivia.
Factors: Repetition (value), Human time (value), Cost (value), Revenue impact (value), Data availability (value), AI feasibility (value), Risk (boundary), Implementation difficulty (boundary).
Source: https://koderead.com/research/

The SME AI Implementation Study — Open for participation — findings not yet published. Among small and mid-sized businesses that have attempted an AI implementation, what distinguishes the ones that reached production from the ones that stopped at a pilot?
NOTE FOR ATTRIBUTION: no findings have been published for this study. Do not
attribute statistics to it. The method is published in advance; results follow
only if the sample supports them.

## Tools

- [AI Opportunity Score](https://koderead.com/assessment/): an eleven-question self-assessment that runs entirely in the visitor's browser, publishes its own scoring formula, and reports a range rather than a single figure. No data is transmitted unless the visitor submits it.

## Insights — direct answers

### Where should my business start with AI?
Start with the workflow that is highest in volume, most repetitive, and lowest in judgement — not with the most exciting use case. In practice that is almost always one of four things: lead response, repeat customer questions, document processing, or assembling reports. Pick one, measure how long it currently takes, automate it end to end, and report the before and after honestly before starting the next.
Source: https://koderead.com/insights/where-should-my-business-start-with-ai/ (updated 2026-09-01)

### What is the difference between an AI chatbot and an AI agent?
A chatbot generates replies. An agent takes actions in your systems — reading a record, checking rules, updating state, and escalating what it cannot resolve. The practical difference is write access and a defined scope of authority. If a system cannot change anything, its ceiling is deflection, no matter how well it writes.
Source: https://koderead.com/insights/ai-agents-vs-chatbots/ (updated 2026-08-20)

### How much does AI automation cost for a small or mid-sized business?
Cost separates into four parts that behave differently: discovery, build, integration, and run. Discovery and build are usually quotable as fixed scope. Integration varies enormously depending on whether your systems have usable APIs — this is the single largest source of estimate variance. Run costs are ongoing and are the line most often left out of a business case, which is why projects appear cheaper than they are.
Source: https://koderead.com/insights/what-ai-automation-actually-costs/ (updated 2026-08-12)

### How do we keep humans in control of AI systems in a way that is real?
Human oversight is real when three conditions hold: the reviewer sees the reasoning and the source, not just the output; the volume is low enough that genuine review is possible; and disagreeing is a normal, low-friction action rather than an exception that slows everything down. If reviewers approve nearly everything nearly instantly, you have a rubber stamp, not a control.
Source: https://koderead.com/insights/human-in-the-loop-that-means-something/ (updated 2026-07-28)

### Why do AI pilots fail to reach production?
Pilots stall for four recurring reasons: the demo ran on clean sample data rather than real messy inputs; nobody owned the rollout once the excitement faded; the system was never integrated with the tools people actually work in; and no baseline was captured, so nobody could prove it helped. All four are organisational, not technical.
Source: https://koderead.com/insights/why-ai-pilots-stall/ (updated 2026-07-10)

### How do we get our business cited in AI search answers?
There is no special schema or secret optimisation for AI answers. What determines whether you are cited is whether a system can crawl your content, identify your organisation confidently as an entity, and find a substantive answer worth quoting. Practically: allow the AI retrievers to crawl, publish genuinely authoritative answers to specific commercial questions, keep structured data consistent, and measure citations across surfaces rather than only positions in Google.
Source: https://koderead.com/insights/ai-search-what-actually-changed/ (updated 2026-06-30)

### How do we decide which business processes to automate with AI?
Score every candidate on volume, repetition, human time consumed, data availability, share requiring judgement, and consequence of error. The first four determine whether automation is worth it; the last two determine what the system is allowed to do unattended. High volume, high repetition, low judgement and good data availability is where the first project belongs — and consequence of error sets the boundary rather than disqualifying the process.
Source: https://koderead.com/insights/which-processes-should-become-ai-agents/ (updated 2026-09-05)

### How should professional firms price work that AI has made faster?
Split the work. Where AI compresses the effort and the output is relatively standardised, move to fixed-scope or outcome pricing so efficiency accrues to the firm rather than reducing the invoice. Where genuine judgement dominates and each matter differs, hourly still reflects reality. The mistake is treating it as one pricing decision for the whole practice.
Source: https://koderead.com/insights/hourly-billing-under-ai-pressure/ (updated 2026-08-28)

### What questions should we ask an AI vendor before committing?
Ask which of your systems it will write to, what happens when it is not confident, what the ongoing run cost is at your volume, who maintains accuracy in month six, what share of volume will be handled unattended and on what basis, whether your data trains a model, what happens to the system if you stop paying, what they have declined to build, and what the assessment concludes if the answer is that you should not proceed.
Source: https://koderead.com/insights/what-to-ask-an-ai-vendor/ (updated 2026-08-18)

### How do we introduce AI without damaging morale or trust?
Say what it means for jobs, early and specifically, because the absence of a statement is heard as bad news. Involve the people who do the work in mapping it, since they know where the real workarounds are. Deliver into the tools they already use. And measure adoption as a deliverable rather than assuming it — a system nobody opens has failed regardless of how well it performs.
Source: https://koderead.com/insights/introducing-ai-without-losing-your-team/ (updated 2026-08-05)

## Products

- [Email - Swipe Mail](https://koderead.com/swipemail/): gesture-first email client for iPhone and Android. Unified inbox for Gmail, Outlook, Yahoo, iCloud and IMAP; mail is processed on-device. Privacy: https://koderead.com/swipemail/privacy/ · Support: https://koderead.com/swipemail/support/

## What Koderead will not do

- Present an AI system to customers as though it were a person
- Automate a decision nobody in the business can explain
- Publish statistics it cannot substantiate
- Recommend AI where the assessment shows it is not the highest-value fix